Summary of Robustness Of Deep Neural Networks For Micro-doppler Radar Classification, by Mikolaj Czerkawski and Carmine Clemente and Craig Michie and Christos Tachtatzis
Robustness of Deep Neural Networks for Micro-Doppler Radar Classification
by Mikolaj Czerkawski, Carmine Clemente, Craig Michie, Christos Tachtatzis
First submitted to arxiv on: 21 Feb 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG); Signal Processing (eess.SP)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This research paper explores the limitations of deep learning-based radar data processing systems. Two convolutional neural networks were trained and tested on the same dataset, revealing that they are sensitive to subtle changes in the input representation and prone to overfitting. The study shows that this sensitivity is due to the models focusing on features that don’t generalize well. As a potential solution, training the models on adversarial examples and temporally augmented samples can improve their robustness. Additionally, the paper finds that using a cadence-velocity diagram representation instead of Doppler-time can make the models more resistant to attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research shows how deep learning-based radar systems can be limited by their ability to learn features that don’t work well in different situations. The study looks at two special types of computer programs called convolutional neural networks and finds that they are easily fooled by small changes in what they’re looking at, which isn’t good for making decisions. To fix this problem, the researchers suggest training these systems on examples that are intentionally made tricky to process. They also find that using a different way of looking at radar data can make the systems less susceptible to these problems. |
Keywords
* Artificial intelligence * Deep learning * Overfitting
